PLASMA

 

最新版本

• Add  LU  inversion functions: PLASMA_zgetri, PLASMA_zgetri_Tile and PLASMA_zgetri_Tile_Async using   the   recursive  parallel  panel implementation of LU factorization.
• The  householder  reduction  trees for QR and LQ factorizations can now  work on general cases and not only on matrices with M multiple of MB.
• Matrices  generation  has been changed in every timing, testing and example  files to use a parallel initialization generating a better distribution  of  the data on the architecture, especially for Tile interface. “numactl” is not required anymore.
• Timing  routines  can  now generate DAGs with the --dag option, and traces with --trace option if EZTRACE is present.
  

 Purpose of PLASMA（来自于软件包的readme）

   The  main  purpose of PLASMA is to address the performance shortcomings
   of the [2]LAPACK and [3]ScaLAPACK libraries on multicore processors and
   multi-socket  systems  of  multicore  processors and their inability to
   efficiently  utilize  accelerators  such  as  Graphics Processing Units
   (GPUs).  PLASMA  provides  routines  to  solve dense general systems of
   linear   equations,  symmetric  positive  definite  systems  of  linear
   equations and linear least squares problems, using LU, Cholesky, QR and
   LQ factorizations. Real arithmetic and complex arithmetic are supported
   in both single precision and double precision.

   PLASMA has been designed to supercede LAPACK and ScaLAPACK, principally
   by restructuring the software to achieve much greater efficiency, where
   possible,  on  modern  computers  based on multicore processors. PLASMA
   also  relies  on new or improved algorithms. Currently, however, PLASMA
   does  not  serve  as  a  complete  replacement of LAPACK due to limited
   functionality.  Specifically, PLASMA does not support band matrices and
   does  not  solve  eigenvalue  and singular value problems. Also, PLASMA
   does   not   replace  ScaLAPACK  as  software  for  distributed  memory

computers, since it only supports shared-memory machines
                                                           
 PLASMA的安装 

1准备阶段

安装BLAS

安装CBLAS

安装LAPACK

2修改make.inc

进行编译

3测试

timing.c是时间和flops函数

问题：

1测试plasma时候怎么把blas设置为1？